All-Optical Flip-Flops Based on Semiconductor Technologies

Abstract

1

Full text

Page 1

All-optical flip-flops based on semiconductor technologies347
x

All-optical flip-flops based on
semiconductor technologies

Antonella Bogoni1, Gianluca Berrettini2, Paolo Ghelfi1,
Antonio Malacarne2, Gianluca Meloni1, Luca Potì1 and Jing Wang3
1Consorzio Nazionale Interuniversitario per le Telecomunicazioni (CNIT), Pisa
Italy
2 Scuola Superiore Sant’Anna, Pisa
Italy
3Department of Electronic Engineering, Tsinghua University, Beijing
China

1. Introduction

Optical technologies represent the main bet for future communication systems. Among the
others, digital subsystems for optical processing are of great interest thanks to their intrinsic
properties in terms of bandwidth, transparency, immunity to the electromagnetic
interference, cost, power consumption, as well as robustness in hostile environment. Key
basic functions are represented by logic gate, logic function, flip-flop memories, optical
random access memories, etc.. Research in this field is in its very early stages even if some
interesting techniques have been already theoretically addressed and experimentally
demonstrated. Here we review the state of the art for all-optical flip-flop based on
semiconductor technologies: best result will be highlighted in terms of transition speed,
switching energy, complexity and power consumption; we will then discuss some new
achievement we have recently reached.
All-optical packet switching seems to be the most promising way to take advantage of fiber
bandwidth to increase routers forwarding capacity, being able to achieve very high data rate
operations. All-optical flip-flops have been widely investigated mainly because they can be
exploited in all-optical packet switches, where switching, routing and forwarding are
directly carried out in the optical domain. Some examples concerning optical packet
switches are shown in (Dorren et al., 2003; Liu et al., 2005; Bogoni et al., 2007; Herrera et al.,
2007), where an optical flip-flop stores the switch control information and drives the
switching operation. Former solutions for all-optical flip-flops have been demonstrated
exploiting discrete devices (Dorren et al., 2003) or Erbium-doped fiber properties (Malacarne
et al., 2007) which suffer from slow switching times and high set/reset input powers.
Several integrated or integrable solutions (Hill et al., 2004; Liu et al., 2006) present a
switching energy in the fJ range and switching times of tens of ps at the expenses of poor
contrast ratios. On the other hand in (Hill et al., 2005) an integrated scheme exhibiting a very
high contrast ratio value but with transition times in the ns range is reported. In any case a

15

Page 2

Semiconductor Technologies 348

trade off between contrast ratio and edges speed must be found as a function of the flip-flop
application. Micro-resonators-based bistable element has been demonstrated (Van et al.,
2002) presenting high optical operating power, pJ switching energies and microsecond
switching times, theoretically reducible down to the order of tens of ps. Making a
comparison with electronics, recent large-scale integration (LSI) circuits (Keyes, 2001) show
switching energies of 1fJ even though with slower switching speeds. In (Dorren et al., 2003),
a solution based on coupled ring lasers is proposed. This solution offers a certain number of
advantages: it can provide high contrast ratios between states; there is no difference in the
mechanisms for switching from state 1 to state 2 and vice-versa, making symmetric set and
reset operations; it presents a large input light wavelength range and a controllable
switching threshold. Moreover, considering an integrated version of this kind of flip-flop,
through numerical analysis a switching energy in fJ range has been demonstrated.
Here we will describe the above mentioned solutions underlining the main benefits,
drawback, limitation and perspectives. We will then present our activities on clocked flip-
flops, and an example of their use in an all-optical counter. Finally, we will present an SOA-
based flip-flop which is able to switch with very short rising and falling edges, and we use it
in a realistic switching operation. Integrability of our solutions is also discussed.

2. State of the art

One of the simplest way that was originally proposed to implement an optical flip-flop
includes two coupled lasers (Hill et al., 2001), as depicted in Fig. 1 (a). The system can have
two stable states. In state 1, light from laser 1 suppresses lasing in laser 2. In this state, the
optical flip-flop memory emits CW light at wavelength 1. Conversely, in state 2, light from
laser 2 suppresses lasing in laser 1, and the optical flip-flop memory emits CW light at
wavelength 2. To change states, lasing in the dominant laser can be inhibited by injecting
external light with a different wavelength and opportune power. The output pulse of an
optical header processor can be used to set the optical flip-flop memory into the desired
wavelength. From the theory it also follows that laser driving currents and coupling
coefficient determines the required switching light power.
This flip-flop has also been implemented in a ring configuration based on Semiconductor
Optical Amplifiers (SOA), as shown in Fig. 1 (b) (Dorren et al., 2003). Two SOAs act as the
lasers gain media. Fabry–Pérot filters (FPF) with a bandwidth of 0.18nm have been used as
wavelength selective elements. Optical pulses were used to set and reset the flip-flop. The
optical spectrum of the flip-flops’ output states is shown in Fig. 2.
The switching time between the two lasing modes is inversely proportional to the length of
the laser cavities. Thus, in order to allow switching times in the range of picoseconds, an
integrated solution has to be adopted. This was realized in (Hill et al., 2004), where a
photonic flip-flop based on two coupled micro-ring lasers with dimensions of 20x40 m2
was reported, exhibiting a switching time of 18ps and a switching energy of a few fJ.
The micro-ring lasers were fabricated in active areas of the integrated circuit containing bulk
1.55nm bandgap InGaAsP in the light guiding layer. Separate electrical contacts allowed
each laser’s wavelength to be individually tuned by adjusting the laser current. Passive
waveguides connected the micro-ring lasers to the integrated circuit edges (Fig. 3). Micro-
ring lasers typically have two inherent lasing modes; laser light traveling in the clockwise
(CW) direction, and laser light in the anticlockwise (ACW) direction.

Page 3

All-optical flip-flops based on semiconductor technologies349


(a) (b)
Fig. 1. (a): Arrangement of two coupled identical lasing cavities forming a flip-flop,
showing the two possible states. (b): Implementation of the optical flip-flop memory


Fig. 2. Spectral output of two states of the optical flip-flop memory.


Fig. 3. Two micro-ring lasers coupled via a waveguide to form an optical flip-flop.

In state A, CW light from laser A is injected via the waveguide into laser B. The light from
laser A will undergo significant resonant amplification in laser B if the resonant frequencies
of the two laser cavities are close. This injected light competes with the laser B self-
oscillations for available power from the laser gain medium. If sufficient light is injected into
laser B, then the laser B gain will be decreased below threshold. This extinguishes the laser B
self-oscillation, and laser A captures or injection-locks (Buczek et al., 1971) laser B, forcing
light to circulate only in the CW direction. To set the system in one state or another, light

Page 4

Semiconductor Technologies350

close to the lasing wavelength and polarization can be injected into the waveguide
connecting the lasers. This light will set both lasers simultaneously lasing in either the CW
or ACW direction. The different states can be distinguished by the different power levels at
the two outputs. The power level at the output associated with the locked laser will be three
times that of the other output. Additionally, the lasing wavelengths of the lasers may be
different, allowing the states to be distinguished by the wavelength of the light output.
Another scheme recently proposed (Malacarne et al., 2007) exploits absorption and
fluorescence of few meters of erbium–ytterbium (Er–Yb)-doped fiber. This solution suffers
from slow switching times and high set/reset input powers, and since it doesn’t exploit
semiconductor devices, it will not be studied in depth here.
In (Liu et al., 2006) a solution that offer the advantage of being fully packaged, was
presented. It is based on an hybrid integrated circuit consisting of two coupled Mach-
Zehnder interferometers (MZIs), each having one SOA in one arm. The schematic of the
circuit is shown in Fig. 4.


Fig. 4. Schematic diagram of optical flip-flop memory proposed in (Liu et al., 2006).

Each MZI (MZI 1 and MZI 2 in the figure) has an SOA in one arm. A laser emits a
continuous-wave (CW) bias light at wavelength 1 that is fed into MZI 1. The MZI 1 output
is sent into MZI 2, which has the same structure, but biased by a CW light with a different
wavelength, 2. The system has two possible states: in state 1, the MZI 1 output suppresses
output from MZI 2, so 1 dominates the output; in state 2, the MZI 2 output suppresses
output from MZI 1, and then 2 is dominant. When the CW light with 1 is injected into MZI
1, MZI 1 is biased in such a way that the light out of MZI 1 goes mostly into the low branch
of the 50/50 coupler output. This light then flows into MZI 2 via the 50/50 coupler in MZI 2,
and affects the gain and phase shift for light propagating through it. The MZI 1 light
perturbs the SOA 2 properties so that the CW bias 2 light (2) propagating through SOA 2
and phase shifter 2 goes mostly into the top output of the 50/50 coupler in MZI 2. Then the
CW bias 2 light (1) does not travel into the MZI 1, and does not affect the properties of SOA
1. Actually, the MZI 1 output suppresses output from MZI 2. The states of the system can be
switched by sending a light pulse (via Set or Reset port) into the MZI that is currently
dominant. This light will switch the MZI output away from suppressing the other MZI,
allowing the other MZI then to become dominant.
An optical flip-flop based on two-mode bistability in a multimode interference bistable laser
diode (MMI-BLD) has also been reported (Takenaka et al., 2005). A schematic view of the
MMI-BLD is shown in Fig. 5 (a). All waveguides including the 2x2 MMI coupler consist of
active materials. Saturable absorbers are located at the end of the output ports to obtain
hysteresis. The 2x2 MMI is designed as a cross coupler, so that only two cross-coupled
lasing modes can exist as illustrated in the insets of Fig. 5 (a). Two-mode bistability between

Page 5

All-optical flip-flops based on semiconductor technologies351

these two lasing modes will occur due to cross gain saturation and the saturable absorbers if
the injection current is within the hysteresis loop (Takenaka & Nakano, 2003).

(a) (b)
Fig. 5. (a): Schematic view of the MMI-BLD. Two cross-coupled lasing modes are
illustrated in the insets. (b): All-optical flip-flop operation of the MMI-BLD.

A set signal injected into the set port saturates the absorption to Mode 1, causing Mode1 to
start lasing. At the same time, cross-gain saturation and the absorption to Mode 2 by the
saturable absorber suppress Mode 2. In a similar manner, a reset signal switches the lasing
mode from Mode 1 to Mode 2. Therefore, all-optical flip-flop operation is achievable with
the MMI-BLD, because external light injection to each input port will select the mode to lase.
The corresponding operation, showing the optical power at one of the waveguide output
when set and reset pulses are applied is depicted in Fig. 5 (b).
In (Huybrechts et al., 2008) a single DFB laser diode has been used to realize a flip flop. A
DFB laser injected with CW light shows two different stable states: one in which the laser is
lasing and another one where it is switched off. When the laser is lasing, the gain will be
clamped and relatively small. Therefore, the injected light experiences only a small
amplification and has almost no influence on the laser light. In the second state, the laser is
switched off and the injected light experiences a high amplification. This results in a rising
power progression throughout the cavity and therefore a non-uniform distribution of the
carriers, known as spatial hole burning. This will affect the refractive index, leading to a
distortion of the Bragg reflections in the laser diode. The losses inside the cavity will become
higher and the threshold for lasing will rise. Eventually the laser will stay switched off. The
two states are equally possible for a range of input powers of the injected light and this gives
a bistability in the lasing power (Fig. 6 (a)). This bistability can be exploited to obtain flip-
flop operation by injecting short optical pulses: a pulse injected at the same side as the CW
light will move the DFB laser out of the hysteresis curve and will switch off the laser; to
switch the laser on again, a pulse is injected from the other side, since this will restore the
uniformity of the carrier distribution. In the experiment, the set and reset pulses were
obtained from an ultra-short pulse source generating 7ps-long pulses. The obtained results

Page 6

Semiconductor Technologies352

are depicted in Fig. 6 (b). The set-pulses have an energy of 75fJ and the reset-pulses 190 fJ.
The repetition rate is 1.25GHz and the switch-on time is 75ps. An almost immediate switch-
off time of 20ps has been obtained, which corresponds with the resolution of the optical
scope.


(a) (b)
Fig. 6. (a): Bistability of an injected DFB laser as a function of the injected power. (b): Results.




(a) (b)
Fig. 7. (a): Operation principle of the monolithic semiconductor ring laser. (b): Results.

As discussed previously, integrable solutions are preferred since they would allow high-
density packaging, with the possibility of reducing costs, power consumption, and
operation speed. To achieve these results, researchers are investigating novel technologies in
order to reduce as much as possible device dimensions. A possible solution towards this
direction is the use of a monolithic semiconductor micro-ring laser (Trita et al., 2009) which
shows an intrinsic and robust directional bistability between its CW and ACW propagating
modes. If the ring laser is correctly set, injecting a laser pulse in one direction makes the
laser emit in that direction (Fig. 7 (a)). Experiments show a switching time of about 20ps for
both rising and falling edges, with set/reset pulses of 5ps and 150fJ energy.
Another promising technology is nano-photonics, exploited in the realization of photonic
crystals (PCs) and quantum dots (QDs). By combining these technologies one could take

Page 7

All-optical flip-flops based on semiconductor technologies353

advantage of both the band-gap effect and the highly dispersive property of PCs, and the
high-density of state and high nonlinear property of QDs.


Fig. 8. Schematic diagram of the PC-FF.

A Mach Zehnder-type all-optical flip-flop developed by combining GaAs-based two-
dimensional photonic crystal (2DPC) slab waveguides and InAs-based optical nonlinear
QDs has been proposed in (Azakawa, 2007). The photonic crystal-based flip-flop (PC-FF)
schematic is shown in Fig. 8, and is based on two photonic-crystal-based Symmetric Mach
Zehnder (PC-SMZ) switches. The principle of the PC-SMZ is based on the time-differential
phase modulation caused by the nonlinear-induced refractive index change in one arm of
the two interferometers. 2DPC waveguides are composed of single missing line defects,
while nonlinear-induced phase shift arms are selectively embedded with QDs. The
mechanism of the third-order nonlinear property is an absorption saturation of the QD
caused by a control (pump) pulse. A resultant refractive index change produces a phase
shift for the signal (probe) pulse. A wavelength of the control pulse is set to the absorption
peak of the QD, while a wavelength of the signal pulse is set in the high transmission range
in the 2DPC waveguide with the QD. A single PC-SMZ switch would operate as a pseudo-
flip-flop, meaning that the on-state is limited by the carrier relaxation time in the nonlinear
material (~ 100ps in the experiment). In order to change the pseudo FF into the normal FF
operation, the scheme of Fig. 8 was proposed. An output signal of the PC-SMZ impinges
into an optical AND element (which is another PC-SMZ switch) via a feedback loop, where
another input pulse, i.e., a clock pulse impinges. An output of the AND element is combined
to the set pulse, as shown in the figure. The clock pulse serves as a refresh pulse to expand
the on-state period against the relaxation of the carrier, while the feedback signal restricts
the clock pulse to be controlled by the set and reset pulses. The feasibility of this idea has
been verified only by computer simulation.

3. Flip-flops based on coupled SOA ring lasers: advantages and limitations

In order to investigate advantages and drawbacks of SOA-based solution we consider the
setup shown in Fig. 9. The flip-flop consists of two coupled ring lasers emitting at two
different wavelengths (λ1=1550nm and λ2=1560nm). In each ring, an SOA acts as the gain
element, a 0.25nm band-pass filter (BPF) is used to as select the wavelength, and an isolator
makes the light propagation unidirectional. Both the SOAs are polarization insensitive

Page 8

Semiconductor Technologies354

Multi-Quantum Well (MQW) structures with a small-signal gain of 31dB, saturation power
of 13dBm and Amplified Spontaneous Emission (ASE) noise peak at 1547nm.

1

2


Fig. 9. Experimental Setup of the all-optical flip-flop based on SOAs.

555
-65-65 -65
-55-55 -55
-45-45 -45
-35-35-35
-25-25-25
-15-15-15
-5-5-5
154515451545 155015501550155515551555156015601560 156515651565
Power (dBm)
Wavelength (nm) Wavelength (nm)Wavelength (nm)
laser 1 laser 1laser 1laser 2 laser 2laser 2
CR>40dBCR>40dBCR>40dB
CR=50dB CR=50dBCR=50dB
-16-16-16-14-14-14-12-12-12-10-10-10-8-8-8-6-6-6-4-4-4-2-2-2000222444
-55-55-55
-50-50-50
-45-45-45
-40-40-40
-35-35-35
-30-30-30
-25-25-25
-20-20-20
-15-15-15
-10-10-10
static switching, in case of external CW light injected into laser 1
CW light injected into ring 1
static switching, in case of external CW light injected into laser 1
CW light injected into ring 1
static switching, in case of external CW light injected into laser 1
CW light injected into ring 1
P injected (dBm)P injected (dBm)P injected (dBm)
Pout (dBm)


laser 1
laser 2laser 2laser 2
-16-16-16-14 -14-14-12 -12-12-10 -10-10-8 -8-8-6 -6-6-4-4-4-2 -2-2000222444
-55 -55-55
-50 -50-50
-45 -45-45
-40 -40-40
-35 -35-35
-30 -30-30
-25 -25-25
-20 -20-20
-15 -15-15
-10 -10-10
static switching, in case of external CW light injected into laser 2
CW light injected into ring 2
static switching, in case of external CW light injected into laser 2
CW light injected into ring 2
static switching, in case of external CW light injected into laser 2
CW light injected into ring 2
P injected (dBm) P injected (dBm)P injected (dBm)
Pout (dBm)


laser 1
laser 2laser 2laser 2
CR=40dBCR=40dBCR=40dB
CR>40dBCR>40dBCR>40dB
CR>40dBCR>40dBCR>40dB
CR=40dBCR=40dBCR=40dB
Power (dBm)Power (dBm)
Pout (dBm)
laser 1laser 1
Pout (dBm)
laser 1laser 1
Pout (dBm)
Pout (dBm)

Fig. 10. Top: optical spectra of the two states; Bottom: output power of lasers versus input
power injected into cavity 1 ( left) and into cavity 2 (right).

The system can have two states. In “state 1”, light from ring 1 suppresses lasing in ring 2,
reaching cavity 2 through the 50/50 coupler and saturating the SOA 2 gain. In this state, the

Page 9

All-optical flip-flops based on semiconductor technologies355

optical flip-flop output 1 emits CW light at wavelength λ1.In “state 2” light from ring 2
suppresses lasing in ring 1 (saturating SOA 1 gain), and output 2 emits CW light at
wavelength λ2. To dynamically change state, lasing in the dominant cavity can be switched
off by injecting external pulsed light with a wavelength different from λ1 and λ2
(λIN=1554.5nm). In Fig. 10 experimental measurements of the two states optical spectra are
investigated and a graph of the output power of both the ring lasers, versus the CW input
power injected into each cavity is reported. The output contrast ratios are higher than 40dB.

1
051015202530354045
0
0.5
set
0510152025303540 45
0
0.5
1
reset
05 1015202530354045
0
0.5
1
ring 1
0510152025303540 45
0
0.5
1
time (us)
ring 2

Fig. 11. Experimental results of the all-optical flip-flop output.

(a)(a)
(b)(b)
(c)(c)(d)(d)

Fig. 12. Measured (a)-(b) and simulated (c)-(d) behavior of the flip-flop output edges.

By injecting two regular sequences of pulses into the set and reset ports, we demonstrate the
dynamic flip-flop operation shown in Fig. 11. We experimentally observed that the flip-flop
falling time only depends on the edge time of control pulses (5ns in this section), while the
rising time is determined by the cavity length and by the length of the fiber between the two
SOAs. In our setup, each ring has a cavity length of 20m corresponding to a round-trip time

Page 10

Semiconductor Technologies356

of about 100ns. Experimental measurements (Fig. 12 (a)) show that the building-up process
of one state takes place step by step and each step corresponds to a cavity round-trip time
equal to 100ns. The total rising edge behavior lasts several hundreds of ns. The experimental
falling edge behavior is shown in Fig. 12 (b), with a transition time of 5ns, equal to the input
pulse edge.
Dynamics behavior of the two SOA-based coupled lasing cavities has been analyzed
through simulations as well, whose details can be found in (Barman et al., 2007). Assuming
the same parameters of the experimental setup (cavity length and cavity loss, injected pulses
edge time and average power), as can be observed in Fig. 12 (c)-(d), simulation results for
rising and falling edges are in good agreement with experimental measurements,
confirming the step behavior of the rising edge and at the same time a falling edge as fast as
the input pulse edge. We also simulated an integrated version of this flip-flop, considering
2mm cavity length and 0.5mm SOA length. Results predict 12ps falling time and ~40ps
rising time with injected input pulsewidth of 12ps and pulse energy of 15.6fJ, comparable
with the results of one of the latest optical flip-flop integrated version (Hill et al., 2004).

4. SOA-based clocked flip-flops

Most of the all-optical flip-flops proposed in literature are non-clocked devices, whose
output changes immediately following the set/reset signals, thus they are also referred to as
Set-Reset (SR) latch. As a digital device that temporarily memorizes the past input signal
and processes it with current inputs, optical flip-flop is expected to be synchronized with a
system clock, and to work in a timely programmed mode. Moreover, in some complicated
optical computing applications such as optical shift registers or counters, various types of
clocked flip-flops are necessary, such as SR, D, T, and JK flip-flops.
Starting from the basic structure defined in the previous paragraph, here we show clocked
all-optical flip-flops including SR, D, T, and JK types, exploiting also AND logic gates based
on nonlinear effects in SOA (Wang et al., 2009, a).

4.1 Clocked SR flip-flop
The characteristic table of the set/reset (SR) flip-flop is shown in Fig. 13 (a). If S=R=0, the flip-flop
remains at its previous state; if S=1 R=0, it is set to “state 1”; if S=0 R=1, it is set to “state 0”. S=R=1
is forbidden since the flip-flop is unstable in this case. The setup of clocked SR flip-flop is shown
in Fig. 13 (b): it consists of two AND gates and one SR latch. “AND 1” and “AND 2” perform
AND function between the clock pulse and S and R, respectively. The outputs of “AND 1” and
“AND 2” are connected to the “Set” and “Reset” ports of the latch respectively. The operation
principle of this clocked flip-flop is shown in Fig. 13 (c): when a clock pulse comes, if S=R=0 it can
not pass through either “AND 1” or “AND 2”, so “Set” and “Reset” ports receive no pulse and
the latch maintains its previous state (Qnext=Q); if S=1 R=0, the clock pulse can pass through
“AND 1” but is blocked by “AND 2”, so only “Set” receives a pulse and the latch is set to “state
1” (Qnext=1); if S=0 R=1, the clock pulse can pass through “AND 2” but is blocked by “AND 1”, so
the latch is set to “state 0” (Qnext=0). S=R=1 is forbidden since the latch is unstable when “Set”
and “Reset” receive pulses simultaneously. The flip-flop is clocked because it only changes state
when a clock pulse comes, according to the S and R values at that time. S and R values at any
other time are ignored.

Page 11

All-optical flip-flops based on semiconductor technologies357

ForbiddenForbidden N/AN/A1111
Reset Reset001100
Set Set110011
Hold state Hold stateQQ0000
CommentCommentQnext
Qnext
RRSS

(a) (b) (c)
Fig. 13. Clocked SR flip-flop: (a) characteristic table; (b) logic circuits; (c) working principle.


Fig. 14. Clocked SR flip-flop operation.

In Fig. 14 the experimental operation of the clocked SR flip-flop is reported. The clock pulse
has a repetition rate of 200kHz with a pulse-width of 1μs. S and R signals also have a pulse-
width of 1μs but at a repetition rate of 50kHz, synchronized with the clock. The wavelengths
of clock, S and R are λCLK=1554.1nm, λS=1552.5nm and λR=1550.5nm respectively, and the
outputs of “AND 1” and “AND 2” are at λ1=2λS-λCLK=1550.9nm and λ2=2λR-λCLK=1546.9nm.
The flip-flop only responses to the S and R values when a clock pulse comes, but ignores the
S and R at any other time, in agreement with Fig. 13 (c).

4.2 Clocked D flip-flop
The characteristic table of D flip-flop is shown in Fig. 15 (a). D represents the data signal. If
D=0, the flip-flop is set to “state 0”; if D=1, the flip-flop is set to “state 1”. The setup of
clocked D flip-flop is shown in Fig. 15 (b): “AND 1” gate performs AND function between
the clock pulse and D, whereas “AND 2” performs AND function between clock and
inverted D. The operation principle of D flip-flop is shown in Fig. 15 (c): when a clock pulse
comes, if D=1 it can pass through “AND 1” but is blocked by “AND 2”, so only “Set” port
receives a pulse and the latch is set to “state 1” (Q=1); similarly if D=0 the clock pulse can

Page 12

Semiconductor Technologies358

pass through “AND 2” but is blocked by “AND 1”, only “Reset” receives a pulse and the
latch is set to “state 0” (Q=0). The flip-flop is clocked because it only changes state when a
clock pulse comes, according to the D values at that time, but ignores D at any other time.
SetSet1111
ResetReset
0000
CommentCommentQnext
Qnext
DD

(a) (b) (c)
Fig. 15. Clocked D flip-flop: (a) characteristic table; (b) logic circuits; (c) working principle.


Fig. 16. Clocked D flip-flop operation.

In Fig. 16 clocked D type flip-flop operation is experimentally demonstrated. The clock
pulse has a repetition rate of 60kHz with a pulsewidth of 1μs; whereas D has a repetition
rate of 100kHz with a pulsewidth of 6μs. The wavelength of clock and D are λCLK=1554.1nm
and λD=1552.5nm respectively, so the output of “AND 1” is at λ1=2λD-λCLK=1550.9nm and
the output of “AND 2” is at λ2=λCLK=1554.1nm, the same with the clock pulse. The flip-flop
only responses to the D values when clock pulses come, and therefore is clocked.

4.3 Clocked T flip-flop
The characteristic table of T flip-flop is shown in Fig. 17 (a). T represents the toggling signal.
If T=0, the flip-flop maintains its previous state; if T=1, the flip-flop changes its state. The
setup of clocked T flip-flop is shown in Fig. 17 (b). Different from SR and D flip-flops, in T
flip-flop, the next state is not determined by external control signals, such as S, R, and D, but
depends on the previous state, so feedback of output Q is used in T flip-flop to carry out the
toggling operation. “AND 1” performs AND function between the clock pulse and T;
whereas “AND 2” performs AND between the output of “AND 1” and the feedback output
Q. “AND 3” carries out AND function between output of “AND 1” and inverted Q. The
operation principle of T flip-flop is shown in Fig. 17 (c): when a clock pulse comes, if T=0 it
is blocked by “AND 1”, neither “Set” nor “Reset” receives pulse, and the latch remains at its
previous state. If T=1, the clock pulse can pass through “AND 1”; then, if Q=1 it can pass

Page 13

All-optical flip-flops based on semiconductor technologies 359

through “AND 2” but is blocked by “AND 3”, so only “Reset” receives a pulse and the latch
toggles to “state 0” (Q=0); if Q=0 the clock pulse can pass through “AND 3” but is blocked
by “AND 2”, only “Set” receives a pulse and the latch toggles to “state 1” (Q=1). In this way,
the flip-flop is triggered by the clock pulse, changing its state if T=1, or maintaining its state
if T=0.

ToggleToggleQQ11
Hold stateHold state
QQ00
CommentCommentQnext
Qnext
TT

T
CLK
CLK∩T
Q
11
10100
CLK∩T∩Q
AND 3
Set
AND 2
Reset
0
AND 1
CLK∩T∩Q

(a) (b) (c)
Fig. 17. Clocked T flip-flop: (a) characteristic Table; (b) logic circuits; (c) working principle.


Fig. 18. Clocked T flip-flop operation.

In Fig. 18 clocked T flip-flop operation is experimentally demonstrated. The clock pulse has
a repetition rate of 60kHz with a pulse-width of 1μs; whereas T has a repetition rate of
100kHz with a pulse-width of 6μs. The wavelength of clock pulse and T are λCLK=1554.1nm
and λT=1552.5nm respectively, so the output of “AND 1” is at λ1=2λT-λCLK=1550.9nm. The
flip-flop output, Q, has a wavelength of λQ=1549.3nm, so the output of “AND 2” is at
λ2=2λQ-λ1=1547.7nm and the output of “AND 3” is at λ3=λ1=1550.9nm, the same with the
output of “AND 1”. The flip-flop is clocked since the state toggling is only triggered when a
clock pulse comes and T=1.


4.4 Clocked JK flip-flop
The characteristic table of JK flip-flop is shown in Fig. 19(a), which could be considered as a

Page 14

Semiconductor Technologies360

combination of SR flip-flop and T flip-flop. Like SR flip-flop, J and K signals are also used as
set and reset signals: J=K=0 makes the flip-flop maintain its previous state; J=1 K=0 sets it to
“state 1”; and J=0 K=1 sets it “state 0”. However, in SR flip-flop, S=R=1 is forbidden, but in
JK flip-flop, J=K=1 is allowed and the flip-flop toggles its state in this condition, like a T flip-
flop.

ToggleToggleQQ1111
ResetReset001100
SetSet110011
Hold stateHold stateQQ0000
CommentCommentQnext
Qnext
KKJJ



CLK
Q10010
CLK∩K∩Q
K
J
01101
AND 2
Reset
AND 1
Set
Q
CLK∩J∩Q

(a) (b) (c)
Fig. 19. Clocked JK flip-flop: (a) characteristic table; (b) logic circuits; (c) working principle.


Fig. 20. Clocked JK flip-flop operation.

The setup of clocked JK flip-flop is shown in Fig. 19(b). The two complementary outputs of
two ring lasers of SR latch are used as Q and inverted Q respectively. “AND 1” carries out
AND function between the clock, J, and inverted Q; whereas “AND 2” carries out AND
between the clock, K, and Q. Similar to SR flip-flop, the JK flip-flop can be set and reset by
external signals, so CLK∩J and CLK∩K are partially carried out in two AND gates.
However, the JK flip-flop can toggle its state like a T flip-flop, so the feedback of Q at
previous state must also be taken into account in the two AND gates. When a clock pulse
comes, if J=K=0 it can not pass through “AND 1” and “AND 2”, so neither “Set” nor “Reset”
receives a pulse, and the latch remains at its previous state. If J=1 K=0, the clock pulse is

Page 15

All-optical flip-flops based on semiconductor technologies361

blocked by “AND 2”, but in “AND 1” there are two possible cases. If Q=1 the clock pulse is
blocked, so “Set” receives no pulse and the latch will remain at “state 1”; otherwise if Q=0
the clock pulse can pass through “AND 1”, and the latch will be set to “state 1”. So in the
case of J=1 K=0, the flip-flop will be set to “state 1” no matter in which state it was.
Similarly, if J=0 K=1, the clock pulse is blocked by “AND 1”. But for “AND 2”, if Q=1 the
clock pulse can pass through, so the latch will be set to “state 0”, otherwise if Q=1 the clock
pulse is blocked and the latch will stay in “state 0”. So the flip-flop will be set to “state 0” no
matter in which state it was. Finally, if J=K=1 we also have to consider two cases of Q. If
Q=1, the clock pulse is blocked by “AND 1” but can pass through “AND 2”, so the latch is
set to “state 0”; otherwise, the clock pulse can pass through “AND 1” but is blocked by
“AND 2”, and the latch is set to “state 1”. In both two cases, the flip-flop changes its state,
which is called state toggling.
In Fig. 20 clocked JK flip-flop operation is experimentally demonstrated. The clock pulse has
a repetition rate of 200kHz and a pulsewidth of 1μs. J and K both quasi-periodic pulse
trains, with repetition rate of 100kHz and pulsewidth of 1μs, synchronized with the clock.
However, in order to realize all four cases of J=K=0, J=1 K=0, J=0 K=1, and J=K=1, in every 4
periods (40μs) of J and K, there is one pulse missed, as shown in Fig.12. It could be observed
that the JK flip-flop operation has a good agreement with Fig. 19(c). The wavelengths of
clock, J, and K are λCLK=1554.1nm, λJ=1552.5nm and λK=1550.5nm respectively, and the
wavelength of Q is λQ=1549.3nm, so the output of “AND 1” is at λ1=2λJ-λCLK=1550.9nm and
the output of “AND 2” is at λ2=2λK-λCLK=1546.9nm.

4.5 Three-state flip-flop
Together with clocked flip-flops, another interesting evolution of the basic flip-flop shown
in paragraph 3 is the upgrade to multi-state flip-flop. A multi-state memory could in fact
extend a 1×2 optical switch to a larger dimension of 1×N, depending on the number of states
of the memory.
The setup of the three-state optical memory is shown in Fig. 21 (Wang et al., 2008, a), which
consists of three coupled SOA fiber ring lasers operating at three different wavelengths. The
memory has three states. In “state 1”, only ring 1 is lasing, whereas ring 2 and ring 3 are
suppressed; the output light of SOA 1 is split by coupler A into two portions: one portion
passes through Path 1 (the dashed red line) and then saturates SOA 3, making ring 3
suppressed; the other portion passes through Path 2 (the dashed green line) and then
saturates SOA 2, making ring 2 suppressed. In “state 1”, the optical memory emits a CW
light at the wavelength of λ1 from output 1 port. Similarly, in “state 2”, only ring 2 is lasing,
and the memory emits a CW light at λ2. Finally in “state 3”, only ring 3 is lasing.
To dynamically change the state, three setting couplers are inserted into the ring cavities,
each corresponding to a particular state. One pulse injected into set 1 port is split to saturate
SOA 3 and SOA 2, and it could not reach SOA 1. Thus ring 2 and ring 3 are both suppressed
while ring 1 could lase; the memory is set to “state 1”. Similarly for set 2 and set 3.